Global Crude Oil Refinery Models:
Parametric and Nonparametric
Methods
Evar Umeozor
September 2023
Doi: 10.30573/KS--2023-MP02
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
1
Acknowledgments
The author thanks all KAPSARC colleagues and support staff for their constructive
discussions on the Oil Value-Chain Analyzer (KOVA) project under which this research
was funded and for their help with the data acquisition procedures, respectively.
About KAPSARC
KAPSARC is an advisory think tank within global energy economics and sustainability
providing advisory services to entities and authorities in the Saudi energy sector
to advance Saudi Arabia's energy sector and inform global policies through
evidence-based advice and applied research.
This publication is also available in Arabic.
Legal Notice
© Copyright 2023 King Abdullah Petroleum Studies and Research Center (“KAPSARC”).
This Document (and any information, data or materials contained therein) (the
“Document”) shall not be used without the proper attribution to KAPSARC. The
Document shall not be reproduced, in whole or in part, without the written permission
of KAPSARC. KAPSARC makes no warranty, representation or undertaking whether
expressed or implied, nor does it assume any legal liability, whether direct or indirect,
or responsibility for the accuracy, completeness, or usefulness of any information that
is contained in the Document. Nothing in the Document constitutes or shall be implied to
constitute advice, recommendation or option. The views and opinions expressed in this
publication are those of the authors and do not necessarily reflect the official views or
position of KAPSARC.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
2
Abstract: There is a dearth of publicly available literature on reliable, evidence-based modeling
representations of global crude oil refineries, which has led to a lack of transparency and consistency
regarding how business and government policy processes are informed in many energy systems and
energy transition modeling scenarios. This paper uses actual field data to identify that 40 unique refinery
configurations account for the global refinery landscape. A machine learning algorithm training extremely
randomized tree regressor (ETR) predictors is applied to develop a nonparametric production model
of refineries. To facilitate computational convenience and suitable deployments in diverse use cases,
a multivariate linear regression (MLR) model of refineries is also presented. For all refined products
captured, the machine learning model demonstrates superior predictive performance, with coefficients of
determination of over 90%, but both the machine learning model and the regression model are useful. Other
performance metrics are assessed for both models.
Keywords: Crude oil, Refinery, Configuration, Machine learning, Regression
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
3
Acronym
Definition
CDU
Crude distillation unit
CS
Condensate splitter
HDTR
Hydrotreater
LPG
Liquified petroleum gas
RFMR
Reformer
DSLR
Desulfurizer
ISMR
Isomerizer
VDU
Vacuum distillation unit
RCC
Resid catalytic cracker
FCC
Fluid catalytic cracker
DH
Distillate hydrocracker
RH
Resid Hydrocracker
CKR
Coker
NAP
Naphtha
GAS
Gasoline
KJF
Kerosene/jet fuel
DGO
Diesel/gas oil
HFO
Heavy fuel oil
COK
Coke
MLR
Multivariate linear regression
ETR
Extremely randomized trees regressor
HSO
Heavy sour crude oil
HSW
Heavy sweet crude oil
MSO
Medium sour crude oil
MSW
Medium sweet crude oil
LSO
Light sour crude oil
LSW
Light sweet crude oil
CAP
Design capacity of refinery
MMb/d
Million barrels per day
Kb/d
Thousand barrels per day
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
4
1. Introduction
C
rude oil refineries produce finished
petroleum products through a series of
processing steps known as refinery unit
operations. The varieties and capacities of process
units, along with their operational characteristics,
differentiate one refinery from another. Over the
years, the refining industry has adopted various
metrics to categorize refineries based on their
capital investment and upgrading capability.
These efforts have resulted in refinery complexity
indicators such as the equivalent distillation
capacity, the Nelson complexity index and the
bottom of the barrel index (Kaiser 2017, Ogjresearch
2022). Although these metrics have helped quantify
the refinery complexity and suitability for various
crude oil feed qualities, they fail to provide relevant
information for refinery modeling applications
(Herce, Martini et al. 2022).
Refinery models are essential for assessing the
economic and energy performance of refineries
and potentially exploiting opportunities to optimize
profitability, efficiency, and policy compliance.
From a process design perspective, these models
facilitate the development of optimal refinery
configurations under prevailing operating and
logistical conditions in local and/or regional
markets. Business and government policymakers
draw insights from modeling and analysis for
investment selection, downstream diversification
and regulatory policy design and compliance.
Identifying the fundamental refinery configuration
is crucial for realizing reliable and deployable
refinery models.
Abella, Motazedi et al. (2015) used a predominantly
North American database to identify 10 unique
refinery configurations for model development. Their
configuration-based modeling approach enabled
the deployment of models for estimating refinery
energy consumption and greenhouse gas (GHG)
emissions. Doing so aided the identification and
exploration of efficiency improvements and emission
reduction opportunities (Motazedi, Abella et al.
2017). Such models are also useful for assessing
crude feed blends and refinery configuration options
to maximize the netback (Nduagu, Umeozor et al.
2018). They have also been extended to quantify
the environmental impact of the refinery life cycle
(Young, Hottle et al. 2019). However, for the effective
application of configuration-based modeling, the
accurate identification of refining process units and
their attributes is crucial.
In 2020, there were 867 global crude oil refineries
with active capacities of 103.32 million barrels
per day (MMb/d), of which approximately 80%
were located outside North America (Platts 2022).
Identifying the unique configurations across
the global crude oil refining landscape would
facilitate individual, national, regional, and global
refinery modeling. Such modeling could promote
various application studies related to, for example,
efficiency improvement, emission reduction, policy
design, netback optimization, and investment
competitiveness analysis. The rest of this paper is
organized as follows: global refinery classification
into design configurations is presented in the next
section. In section 3, modeling methods and the
data for both the parametric and nonparametric
modeling of refinery configurations are presented.
The modeling results are presented in section 4, and
the conclusions and further research opportunities
are discussed in section 5.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
5
2. Global Refinery Configurations
I
n this work, the S&P Global Platts World Refinery
Database (Platts 2022) is used to investigate
global crude oil refinery configurations using
process unit-level data for all operating refineries
during the 2005-2020 period. A clustering technique
is developed based on the availability of major
process units and their capacities relative to other
contiguous units in each refinery plant. For the entire
study period, 40 unique refinery configurations are
identified to represent all possible configurations
that operated each year over the 16-year period.
Essentially, 30 of the identified configurations
have not been previously reported and represent
those found outside the North American region,
particularly in Asia. All identified refinery
configurations are presented in supplementary
information section SI.A.
Figure 1 shows the global refining capacity in
2020 by the identified refinery configurations. The
refinery configurations are classified so that design
complexity increases in relative terms from the
Figure 1: Global crude oil refinery capacities by refinery configuration (2020).
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
6
2. Global Refinery Configurations
lower-numbered simpler refineries to the highernumbered complex refineries. For example, the
simplest refinery – also known as a hydroskimming
refinery – is labeled configuration 0 (Config_0),
while the most complex deep-conversion refinery
is labeled Config_39. At approximately 16.4
MMb/d, the Config_33 refinery has the highest
total capacity in the world, whereas there has
been no existing Config_22 refinery capacity since
2016 due to either shutdowns or retrofitting to a
different configuration class. The last Config_22
refinery was in operation in 2015, with a capacity of
approximately 190.8 Kb/d.
The global refining capacity processes approximately
100 MMb/d of crude oil blends, which are normally
categorized into 6 grades based on their sulfur
content and density. Figure 2 shows the average
global crude oil blend runs by refinery configuration
between 2005 and 2020. Heavier (higher density)
Figure 2: Global crude oil blend runs by refinery configuration (2005-2020 average).
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
7
2. Global Refinery Configurations
and sour (higher sulfur content) crudes, such as
heavy sour crude oil (HSO), heavy sweet crude oil
(HSW), and medium sweet crude oil (MSO), are
mainly processed by higher complexity refineries,
whereas light and medium crude grades (light sweet
crude oil (LSW), medium sweet crude oil (MSW)
and light sour crude oil (LSO)) are mostly refined at
simpler and medium conversion refineries (see Table 1
for a description of crude grades). Furthermore,
crude blends are often matched to each refinery
configuration to maximize the production of the
most desired products at the expense of the
least desirable products. Other reasons for crude
oil blending include satisfying environmental
requirements such as sulfur content limits, exploiting
the price differentials among crude grades in the
market, utilizing locally available crude oil products,
and optimizing the crack spread of refined products.
Currently, the global crude run data for Config_9 and
Config_19 refineries have yet to be captured in the
source database of world refineries.
The combinations of refinery designs and their
operations processing various crude oil blends
yield various arrays of refined product slates.
Figure 3 depicts the global average refined product
Figure 3: Global refined product slates by refinery configuration (2005-2020 average).
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
8
2. Global Refinery Configurations
slates by refinery type between 2005 and 2020. In
contrast to simpler refinery configurations, more
complex refineries achieve deeper conversion of the
crude feed to yield more lighter to middle distillate
products (such as liquefied petroleum gas (LPG),
naphtha (NAP), gasoline (GAS), kerosene/jet fuel
(KJF) and diesel/gas oil (DGO)) and fewer bottom
distillates and residues (such as heavy fuel oil
(HFO)). Doing so also translates into higher yields of
petrochemical feedstocks (such as LPG and NAP).
The composition of products from complex refineries
enables them to achieve better crack spreads
through higher yields of more valuable products and
the processing of heavier and/or sour crude grades,
which are normally sold at discounts compared
to benchmark lighter and/or sweet crude grades.
At the time of this study, the source database had
not reported other bottom products, such as resid
and coke. Additionally, global production data for
Config_9, Config_12, and Config_19 have not yet
been captured in the database.
The configuration-based modeling of refineries has
several benefits. It reduces the modeling task for
large-scale studies encompassing multiregional
and multitemporal analyses. It also reduces the
number of modeling variables required to accurately
specify the appropriate model for a given refinery
configuration. The models utilize the larger volume
of operational datasets from all refineries of similar
configurations, capturing the variabilities in design
capacities and uncertainties in operational schemes.
Given the proprietary nature of refinery data,
information on the applicable refinery configuration
can inform the development of process simulatorbased designs for further assessment studies.
Moreover, configuration-based modeling can drive
best-practice information dissemination among
refining plants operating in similar configurations
through comparative analysis of performances
across operators, countries, and regions. Such
information dissemination can support transparent
and standard refinery energy efficiency and
emission assessments with a view to guiding
toward opportunities to optimize relative operational
performances focused on efficiency, economy, and
emission reduction. Such models are also valuable
for developing realistic energy outlook scenarios
and exploring future potentials for the co-processing
of renewable and hydrocarbon feedstocks in
conventional refinery plants.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
9
3. Material and Methods
A
global crude oil refinery database is used to
assemble design and operational data for
the 2005-2020 period, covering over 11,000
observations of process unit-level capacities, crude
blend runs and product slate yields (Platts 2022).
The dataset is split into two samples for model
training and testing at proportions of 70% and
30%, respectively.
for features ( j 1, , n). An infinite ensemble of
extremely random trees estimates the value of the
target variable (ŷp ) in the following form:
'
'
#! ()
#" ()
𝑦𝑦"! (𝑥𝑥) = ' … ' 𝐼𝐼(#!,…,#") (𝑥𝑥)
'
*⊂{,! ,…,," }
*
𝜆𝜆(#
, 𝑥𝑥+
! ,…,#" )
,# ∈*
(4)
where I i1 ,,in x is the characteristic function of the
hyper-interval, as given by the following:
A nonparametric model is developed by
i
i 1
[ x1i1 , x1 i1 1 [ xn n , xn n [.
i1 , , in implementing the extremely randomized tree
regressor (ETR) method in the Python programming
n
i
i 1
i1 i1 1
[ xn n , xn n [.
i1 , , in 0, , N (5)
language. Details on the ETR algorithm[ xare
1 , x1
available in (Geurts, Ernst et al. 2006). However, in
Xi1 ,,in are real-valued parameters that depend on
this context, for a refinery learning data sample of
examples x i and y i, as well as the parameters for the
size N, the application of the algorithm is expressed
minimum sample size for splitting a node (nmin) and
as follows:
the number of randomly selected features at each
i
i
lsN x , y : i 1, , N (1) node (K). For our refinery model, the set
of features consists of the plant design capacity
i
(CDU basis), the runs of grades of crude oil
where x is the vector of explanatory variables
i
feedstock, as categorized using density and sulfur
(called features) and y is the vector of
contents (see Table 1) into LSW, LSO, MSW,
corresponding output variables (called targets). For
MSO, HSW and HSO, and the types of refinery
the case of n features and p targets, both can be
configurations. Overall, there are 47 features in the
denoted as follows:
model, including the plant capacities, the 6 crude
xi x1i , , xni (2) grades and the 40 refinery configurations. The
ETR algorithm is implemented using the PyCaret
y i y1i , , y ip (3) modeling package in Python (Ali 2020).
The sample values of the j th feature organized in
the order of increasing value are x (j1) , , x (j N )
Table 1. Categories of crude oil grades based on density and sulfur characteristics.
Crude Grade
Density (API)
Sulfur (%)
LSW
> 34
< 0.6
LSO
> 34
> 0.6
MSW
> 25 and < 34
< 0.6
MSO
> 25 and < 34
> 0.6
HSW
< 25
< 0.6
HSO
< 25
> 0.6
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
10
3. Material and Methods
Given the nonparametric nature of the ETR model,
product yields for the array of crude grades
deployment in refinery linear programming (LP)
and product slates. For this reason, a Yeoapplications faces tractability challenges due to
Johnson transformation (Yeo and Johnson
the uncertain number of operations needed to
2000) is applied to the formulation to achieve
obtain the unique optimum solution of the problem.
heteroscedasticity robustness. This transformation
Consequently, a multivariate linear regression
necessitates a reformulation of the model as
(MLR) model that furnishes a model for closedfollows:
form mathematical optimization use cases is
$
$
log$𝑦𝑦&! (𝑥𝑥) + 𝜔𝜔, = 𝛼𝛼",!
+ / 𝛼𝛼%$! ,! 𝑥𝑥% ! + / 𝛼𝛼%,!
log$𝑥𝑥%
also presented. The model is also implemented
%!
%&% !
in the Python programming language using the
'
log yand
'j ' , p x j ' 'j , p log x j (8)
p x 0, p statsmodels library. Using the same dataset
j j'
j'
explanatory variables, the MLR model is estimated
For the transformation parameter value, 1 , the
as follows:
%
model is heteroscedasticity robust. In addition to
𝑦𝑦"! (𝑥𝑥) = 𝛼𝛼",! + ) 𝛼𝛼$,! 𝑥𝑥$ (6) improving homoscedasticity, the transformation
$&'
supports desirable attributes in the lower bound
of the variable space required for mathematical
For the categorical variables (xj') representing
optimization modeling applications. Furthermore,
refinery configurations,
other statistical tests are performed to assess
x j ' j {0,1} (7) normality, autocorrelation, etc. Details on the
statistical tests and results are provided in the
In analyzing the model parameters for Equation 6,
supplementary material (Section SI.C).
heteroscedasticity is observed to be present due
to the nature of the dataset. Significant sparsity
results from actual realizations of crude runs and
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
11
4. Results and Discussion
T
he models are trained with source data
from the Platts World Refinery Database.
The database has approximately 15,000
observations of crude runs and product slates
covering every crude oil refinery plant in the world
that operated between 2005 and 2020. Some
of the observations had incomplete information
about either crude runs or product yields, but all
observations reported individual refinery capacities.
The database was preprocessed to remove the
incomplete entries, bringing the final number
of observations for modeling to approximately
11,000. Model training and testing data sample
splits of 70% and 30%, respectively, are applied.
Figures 4 to 9 show parity plots for all the refined
products predicted with the ETR model.
Figure 4: Parity plot for LPG product prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
12
4. Results and Discussion
Figure 5: Parity plot for NAP product prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Figure 6: Parity plot for GAS product prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
13
4. Results and Discussion
Figure 7: Parity plot for KJF product prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Figure 8: Parity plot for DGO product prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
14
4. Results and Discussion
Figure 9: Parity plot for HFO product prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
The coefficient of determination (R2 ) is above 90%
for all refinery products predicted with the ETR
model, meaning that over 90% of the variation
in the target variables is accounted for by the
model. Using the information from the coefficient of
determination for each target variable and the values
of the coefficients of partial determination, which
measures the proportion of total variation in the
target variable that is explained by each individual
feature (explanatory variable) in the model, it is
possible to evaluate and rank all features based on
their importance in predicting the target. Figures 10
to 15 present the top ten features accounting for the
total variation in the predicted refined products. The
figures confirm that the refinery capacity, crude oil
blend runs, and refinery configurations are essential
variables for quantifying variations in the amounts
of the refined products but with different levels of
importance for each estimation of the volumes of
refined products.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
15
4. Results and Discussion
Figure 10: Feature importance plot showing the top 10 features accounting for the total variation in LPG product
prediction.
Source: KAPSARC Oil Value-Chain Analyzer
Figure 11: Feature importance plot showing the top 10 variables accounting for the total variation in NAP product
prediction.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
16
4. Results and Discussion
Figure 12: Feature importance plot showing the top 10 variables accounting for the total variation in GAS product
prediction.
Source: KAPSARC Oil Value-Chain Analyzer
Figure 13: Feature importance plot showing the top 10 variables accounting for the total variation in KJF product
prediction.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
17
4. Results and Discussion
Figure 14: Feature importance plot showing the top 10 variables accounting for the total variation in DGO product
prediction.
Source: KAPSARC Oil Value-Chain Analyzer
Figure 15: Feature importance plot showing the top 10 variables accounting for the total variation in HFO product
prediction.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
18
4. Results and Discussion
In addition to the coefficient of determination
(R2 ), mean absolute error (MAE) and root mean
square error (RMSE) are two other performance
metrics that are evaluated for both the parametric
and nonparametric models of refined products.
Table 2 presents all performance metrics for the
ETR and MLR models. Evidently, the machine
learning-based, nonparametric ETR model
demonstrates higher predictive performance
for the target refined products. For a given
refinery configuration, the potential sources of
deviations between the predicted and observed
values of refined product flows could be from
the differences in the process subunit-level
designs, process technology, the operational
expertise of plant operators (human factors), and
model suitability. Consequently, 15 other training
algorithms were applied to the datasets, and
none outperformed the ETR model on all target
refined product predictions. On the R2 basis, the
MLR model shows the lowest performance on
NAP prediction, whereas the ETR model has the
lowest performance on HFO prediction. Additional
statistical analysis results for both models are
available in supplementary information sections
SI.B and SI.C.
Refinery configuration-based modeling not only
is useful for estimating the production of refined
products but also can facilitate the quantification of
operational energy consumption, GHG emissions and
economic performance for each type of refinery plant
design or configuration. For a given plant capacity
and configuration, variabilities in the properties and
economics of crude oil blend feedstocks determine
the product yield, product types and the overall
refinery economic margin performance. However,
the subjects of refinery energy requirements and
economics are not the focus of this paper; rather,
they constitute the objectives of future research
leveraging the refinery configuration framework and
models presented in this work. By identifying global
refinery configurations, which encapsulates the
attributes of each refinery plant design in the form of
the configuration variable in the production model, the
approach presented in this paper avoids the need to
explicitly represent and specify design and operating
conditions – such as temperature and pressure – in
the presented production model. The use of actual
field data on global refineries in the model training
ensures that the effects of design and operational
differences across the refinery configuration classes
are captured.
Table 2. Prediction performance metrics for the test of the ETR and MLR models.
Target
Test Performance – ETR Model
Test Performance – MLR Model
2
MAE
RMSE
R2
MAE
RMSE
R
LPG
0.62
1.43
0.95
2.07
3.43
0.77
NAP
1.12
2.72
0.96
4.64
8.53
0.57
GAS
2.53
6.29
0.96
8.32
14.71
0.84
KJF
0.94
2.11
0.97
3.21
5.82
0.83
DGO
2.32
5.34
0.98
6.41
11.43
0.88
HFO
1.93
4.54
0.93
7.46
12.30
0.65
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
19
5. Conclusions
U
ntil now, no comprehensive representation
of the global crude oil refining landscape
has been modeled or reported in the
publicly available literature. In this paper, real
refinery design and operating information from over
800 existing crude oil refineries globally is used
to establish that there are essentially 40 unique
configurations. On this basis, refinery plant data are
used to train a machine learning model to predict
refined product yields given plant capacities,
configurations, and crude blend feedstocks. Due
to the nonparametric form of the machine learning
algorithm and the computational challenges for
refinery LP applications, a parametric model
based on MLR is also proposed. Although the
machine learning model demonstrates superior
predictive performance, the parametric model
is convenient for the closed-form representation
of refinery models as part of a wider energy
system for various purposes or simply for optimal
decision-making.
In addition to operational scheduling and crack
spread optimization use cases, refinery models are
needed in energy outlook and energy transition
scenario analysis. They can provide business and
government decision makers with information on
market opportunities, environmental performance,
and regulatory compliance under specified operating
environments and conditions. This work addresses
the production modeling aspect of refinery models.
Extensions of the modeling approach to energy
requirements and emissions constitute further
research opportunities.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
20
References
Abella, Jessica, Motazedi Kavan, John Guo, and
Joule Bergerson. 2015. “Petroleum Refinery Life Cycle
Inventory Model (PRELIM) PRELIM v1.5.”
Ali, Moez. 2020. “PyCaret: An open source, low-code
machine learning library in Python.”
Geurts, Pierre, Damien Ernst, and Louis Wehenkel. 2006.
“Extremely randomized trees.” Machine Learning 63,
no. 1: 3-42. DOI: https://doi.org/10.1007/
s10994-006-6226-1
Herce, Carlos, Chiara Martini, Marcello Salvio, and Carlo
Toro. 2022. “Energy Performance of Italian Oil Refineries
Based on Mandatory Energy Audits.” Energies 15,
no. 2: 532. DOI: https://doi.org/10.3390/en15020532
Kaiser, Mark. 2017. “A review of refinery complexity
applications.” Petroleum Science 14. DOI: https://doi.
org/10.1007/s12182-016-0137-y
Motazedi, Kavan, Jessica Abella, and Joule Bergerson.
2017. “Techno–Economic Evaluation of Technologies to
Mitigate Greenhouse Gas Emissions at North American
Refineries.” Environmental Science & Technology 51,
no. 3: 1918-1928. DOI: https://doi.org/10.1021/acs.
est.6b04606
Nduagu, Experience, Evar Umeozor, Alpha Sow, and
Dinara Millington. 2018. “An Economic Assessment of the
International Maritime Organization Sulphur Regulations
on Markets for Canadian Crude Oil.” Canadian Energy
Research Institute.
Ogjresearch. 2022. “Worldwide Refinery Survey and
Complexity Analysis.” Oil and Gas Journal. Online
Research Center.
S&P Global Platts. 2022. “Platts World Refinery
Database.” Accessed August 15, 2022. https://www.
spglobal.com/commodityinsights/en/products-services/
oil/platts-world-refinery-database.
Yeo, In-Kwon, and Richard Johnson. 2000. “A New
Family of Power Transformations to Improve Normality or
Symmetry.” Biometrika 87, no. 4: 954-959. DOI: https://
doi.org/10.1093/biomet/87.4.954
Young, Brett, Troy Hottle, Troy Hawkins, Matt Jamieson,
Greg Cooney, Kavan Motazedi and Joule Bergerson.
2019. “Expansion of the Petroleum Refinery Life Cycle
Inventory Model to Support Characterization of a
Full Suite of Commonly Tracked Impact Potentials.”
Environmental Science & Technology 53(4): 2238-2248.
DOI: https://doi.org/10.1021/acs.est.8b05572
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
21
Supplementary Information
Global Crude Oil Refinery Models:
Parametric and Nonparametric
Methods
Evar Umeozora, †
a
King Abdullah Petroleum Studies and Research Center, Riyadh, Saudi Arabia
Email: evar.umeozor@kapsarc.org
Telephone: +966 507 029 871
Section SI.A: Summary process
diagrams of the world’s crude oil
refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
23
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
24
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
25
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
26
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
27
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
28
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
29
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
30
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
31
Section SI.A: Summary process diagrams of the world’s crude oil refineries
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
32
Section SI. B: Error plots for the
extremely randomized tree regressor
(ETR) model
I
n this section, residual analysis is presented
in the following figures for the target refinery
products, including liquefied petroleum gas (LPG),
naphtha (NAP), gasoline (GAS), kerosene/jet
fuel (KJF), gas oil diesel (DGO), and heavy fuel
oil (HFO).
Figure SI.B1: Error analysis for LPG prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
33
Section SI. B: Error plots for the extremely randomized tree regressor (ETR) model
Figure SI.B2: Error analysis for NAP prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Figure SI.B3: Error analysis for GAS prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
34
Section SI. B: Error plots for the extremely randomized tree regressor (ETR) model
Figure SI.B4: Error analysis for KJF prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Figure SI.B5: Error analysis for DGO prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
35
Section SI. B: Error plots for the extremely randomized tree regressor (ETR) model
Figure SI.B6: Error analysis for HFO prediction with the ETR model.
Source: KAPSARC Oil Value-Chain Analyzer
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
36
Section SI.C: Results and statistical
analysis of the multivariate linear
regression (MLR) model
R
egression results and statistical analyses
are presented in the following tables
for refinery products including liquefied
petroleum gas (LPG), naphtha (NAP), gasoline
(GAS), kerosene/jet fuel (KJF), gas oil diesel
(DGO), and heavy fuel oil (HFO).
Table SI.C1. Regression results for LPG prediction with the MLR model.
Dep. Variable:
log(LPG + 1)
Model:
OLS
Method:
Least Squares
R-squared:
0.769
Adj. R-squared:
0.768
F-statistic:
1141.
Prob (F-statistic):
0.00
Log-Likelihood:
-8128.6
No. Observations:
10922
AIC:
1.635e+04
Df Residuals:
10878
BIC:
1.667e+04
Df Model:
43
Covariance Type:
HAC
coef
std err
z
P>|z|
[0.025
0.975]
Intercept
-0.4534
0.014
-33.264
0.000
-0.480
-0.427
Configuration [T.Config_1]
-0.0744
0.011
-6.850
0.000
-0.096
-0.053
Configuration [T.Config_10]
0.1112
0.024
4.636
0.000
0.064
0.158
Configuration [T.Config_11]
0.1840
0.018
10.023
0.000
0.148
0.220
Configuration [T.Config_13]
-0.0244
0.038
-0.643
0.520
-0.099
0.050
Configuration [T.Config_14]
0.0171
0.071
0.240
0.810
-0.122
0.157
Configuration [T.Config_15]
0.0788
0.026
3.027
0.002
0.028
0.130
Configuration [T.Config_16]
0.2024
0.060
3.365
0.001
0.085
0.320
Configuration [T.Config_17]
0.6174
0.065
9.481
0.000
0.490
0.745
Configuration [T.Config_18]
0.1192
0.033
3.651
0.000
0.055
0.183
Configuration [T.Config_2]
0.3015
0.044
6.823
0.000
0.215
0.388
Configuration [T.Config_20]
0.8479
0.097
8.767
0.000
0.658
1.037
Configuration [T.Config_21]
0.7114
0.067
10.647
0.000
0.580
0.842
Configuration [T.Config_22]
-0.5430
0.075
-7.208
0.000
-0.691
-0.395
Configuration [T.Config_23]
0.2038
0.079
2.579
0.010
0.049
0.359
Configuration [T.Config_24]
0.5067
0.034
15.124
0.000
0.441
0.572
Configuration [T.Config_25]
0.2188
0.023
9.671
0.000
0.174
0.263
Configuration [T.Config_26]
-0.0711
0.084
-0.849
0.396
-0.235
0.093
Configuration [T.Config_27]
1.0192
0.047
21.635
0.000
0.927
1.112
Configuration [T.Config_28]
0.3244
0.056
5.826
0.000
0.215
0.434
Configuration [T.Config_29]
0.3810
0.102
3.747
0.000
0.182
0.580
Configuration [T.Config_3]
0.1799
0.038
4.689
0.000
0.105
0.255
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
37
Section SI.C: Results and statistical analysis of the multivariate linear regression (MLR) model
Configuration [T.Config_30]
0.3533
Configuration [T.Config_31]
0.6775
Configuration [T.Config_32]
0.0159
Configuration [T.Config_33]
0.3600
Configuration [T.Config_34]
Configuration [T.Config_35]
Configuration [T.Config_36]
0.045
7.800
0.000
0.265
0.442
0.069
9.802
0.069
0.232
0.000
0.542
0.813
0.817
-0.119
0.151
0.032
11.225
0.000
0.297
0.423
-0.1714
0.193
0.4239
0.115
-0.887
0.375
-0.550
0.207
3.698
0.000
0.199
0.649
1.2426
0.046
26.772
0.000
1.152
1.334
Configuration [T.Config_37]
-0.1259
0.061
-2.053
0.040
-0.246
-0.006
Configuration [T.Config_38]
-0.4856
0.553
-0.878
0.380
-1.570
0.598
Configuration [T.Config_39]
1.3457
0.108
12.497
0.000
1.135
1.557
Configuration [T.Config_4]
-0.0339
0.020
-1.738
0.082
-0.072
0.004
Configuration [T.Config_5]
-0.1077
0.028
-3.835
0.000
-0.163
-0.053
Configuration [T.Config_6]
0.2761
0.043
6.456
0.000
0.192
0.360
Configuration [T.Config_7]
0.4456
0.136
3.285
0.001
0.180
0.712
Configuration [T.Config_8]
0.0296
0.052
0.571
0.568
-0.072
0.131
log(CAP + 1)
0.1762
0.006
27.996
0.000
0.164
0.189
log(LSO + 1)
0.0580
0.006
9.821
0.000
0.046
0.070
log(LSW + 1)
0.1305
0.004
34.701
0.000
0.123
0.138
log(MSO + 1)
0.1542
0.004
39.055
0.000
0.146
0.162
log(MSW + 1)
0.1364
0.005
24.891
0.000
0.126
0.147
log(HSO + 1)
0.0689
0.005
14.163
0.000
0.059
0.078
log(HSW + 1)
0.0636
0.013
4.948
0.000
0.038
0.089
Omnibus:
367.519
Durbin-Watson:
2.012
Prob(Omnibus):
0.000
Jarque-Bera (JB):
831.602
Skew:
-0.193
Prob(JB):
2.63e-181
Kurtosis:
4.296
Cond. No.
258.
Table SI.C2. Regression results for NAP prediction with the MLR model.
Dep. Variable:
log(NAP + 1)
Model:
OLS
Method:
Least Squares
R-squared:
0.570
Adj. R-squared:
0.568
F-statistic:
1641.
Prob (F-statistic):
0.00
Log-Likelihood:
-13259.
No. Observations:
10922
AIC:
2.661e+04
Df Residuals:
10878
BIC:
2.693e+04
Df Model:
43
Covariance Type:
HAC
coef
std err
z
P>|z|
[0.025
0.975]
Intercept
-0.4614
0.022
-21.350
0.000
-0.504
-0.419
Configuration [T.Config_1]
-0.2452
0.024
-10.357
0.000
-0.292
-0.199
Configuration [T.Config_10]
-0.4393
0.051
-8.575
0.000
-0.540
-0.339
Configuration [T.Config_11]
-0.5424
0.033
-16.447
0.000
-0.607
-0.478
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
38
Section SI.C: Results and statistical analysis of the multivariate linear regression (MLR) model
Configuration [T.Config_13]
-0.4970
0.047
-10.511
0.000
-0.590
-0.404
Configuration [T.Config_14]
-0.4682
0.188
-2.488
0.013
-0.837
-0.099
Configuration [T.Config_15]
-0.4474
0.046
-9.636
0.000
-0.538
-0.356
Configuration [T.Config_16]
0.0241
0.060
0.403
0.687
-0.093
0.141
Configuration [T.Config_17]
0.2823
0.067
4.235
0.000
0.152
0.413
Configuration [T.Config_18]
-0.4789
0.053
-9.073
0.000
-0.582
-0.375
Configuration [T.Config_2]
-0.5542
0.049
-11.325
0.000
-0.650
-0.458
Configuration [T.Config_20]
1.1288
0.067
16.775
0.000
0.997
1.261
Configuration [T.Config_21]
0.0852
0.087
0.973
0.330
-0.086
0.257
Configuration [T.Config_22]
1.6348
0.043
37.698
0.000
1.550
1.720
Configuration [T.Config_23]
-0.2702
0.085
-3.172
0.002
-0.437
-0.103
Configuration [T.Config_24]
-0.3665
0.109
-3.355
0.001
-0.581
-0.152
Configuration [T.Config_25]
-0.4261
0.036
-11.886
0.000
-0.496
-0.356
Configuration [T.Config_26]
-0.4739
0.096
-4.919
0.000
-0.663
-0.285
Configuration [T.Config_27]
0.5446
0.065
8.434
0.000
0.418
0.671
Configuration [T.Config_28]
0.8107
0.107
7.585
0.000
0.601
1.020
Configuration [T.Config_29]
1.5174
0.076
19.994
0.000
1.369
1.666
Configuration [T.Config_3]
-0.3403
0.087
-3.894
0.000
-0.512
-0.169
Configuration [T.Config_30]
0.0462
0.059
0.788
0.430
-0.069
0.161
Configuration [T.Config_31]
0.2047
0.083
2.465
0.014
0.042
0.368
Configuration [T.Config_32]
-0.3676
0.076
-4.829
0.000
-0.517
-0.218
Configuration [T.Config_33]
-0.5652
0.053
-10.755
0.000
-0.668
-0.462
Configuration [T.Config_34]
-0.9039
0.287
-3.144
0.002
-1.467
-0.340
Configuration [T.Config_35]
-0.9288
0.184
-5.050
0.000
-1.289
-0.568
Configuration [T.Config_36]
0.9793
0.067
14.663
0.000
0.848
1.110
Configuration [T.Config_37]
0.3372
0.058
5.785
0.000
0.223
0.451
Configuration [T.Config_38]
-0.5872
0.101
-5.841
0.000
-0.784
-0.390
Configuration [T.Config_39]
0.9134
0.118
7.770
0.000
0.683
1.144
Configuration [T.Config_4]
-0.5908
0.043
-13.696
0.000
-0.675
-0.506
Configuration [T.Config_5]
-0.2148
0.056
-3.808
0.000
-0.325
-0.104
Configuration [T.Config_6]
-0.4308
0.079
-5.441
0.000
-0.586
-0.276
Configuration [T.Config_7]
-0.2761
0.094
-2.926
0.003
-0.461
-0.091
Configuration [T.Config_8]
-0.4166
0.064
-6.537
0.000
-0.542
-0.292
log(CAP + 1)
0.3003
0.011
27.165
0.000
0.279
0.322
log(LSO + 1)
0.1199
0.008
14.172
0.000
0.103
0.136
log(LSW + 1)
0.1165
0.006
18.818
0.000
0.104
0.129
log(MSO + 1)
0.1695
0.006
26.372
0.000
0.157
0.182
log(MSW + 1)
0.1144
0.008
14.391
0.000
0.099
0.130
log(HSO + 1)
-0.0206
0.007
-2.751
0.006
-0.035
-0.006
log(HSW + 1)
0.0546
0.020
2.672
0.008
0.015
0.095
Omnibus:
117.178
Durbin-Watson:
2.034
Prob(Omnibus):
0.000
Jarque-Bera (JB):
200.182
Skew:
0.025
Prob(JB):
3.40e-44
Kurtosis:
3.661
Cond. No.
258.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
39
Section SI.C: Results and statistical analysis of the multivariate linear regression (MLR) model
Table SI.C3. Regression results for GAS prediction with the MLR model.
Dep. Variable:
log(GAS + 1)
Model:
OLS
Method:
Least Squares
R-squared:
0.840
Adj. R-squared:
0.839
F-statistic:
2052.
Prob (F-statistic):
0.00
Log-Likelihood:
-11304.
No. Observations:
10922
AIC:
2.270e+04
Df Residuals:
10878
BIC:
2.302e+04
Df Model:
43
Covariance Type:
HAC
coef
std err
z
P>|z|
[0.025
0.975]
Intercept
-0.7067
0.020
-34.841
0.000
-0.747
-0.667
Configuration [T.Config_1]
-0.1880
0.021
-8.933
0.000
-0.229
-0.147
Configuration [T.Config_10]
0.1719
0.036
4.824
0.000
0.102
0.242
Configuration [T.Config_11]
0.2608
0.027
9.535
0.000
0.207
0.314
Configuration [T.Config_13]
-0.5483
0.053
-10.327
0.000
-0.652
-0.444
Configuration [T.Config_14]
0.1048
0.089
1.178
0.239
-0.070
0.279
Configuration [T.Config_15]
-0.0433
0.043
-1.013
0.311
-0.127
0.040
Configuration [T.Config_16]
0.1352
0.059
2.289
0.022
0.019
0.251
Configuration [T.Config_17]
0.1373
0.068
2.031
0.042
0.005
0.270
Configuration [T.Config_18]
0.3600
0.039
9.184
0.000
0.283
0.437
Configuration [T.Config_2]
0.1088
0.068
1.606
0.108
-0.024
0.241
Configuration [T.Config_20]
0.5307
0.139
3.816
0.000
0.258
0.803
Configuration [T.Config_21]
0.2245
0.071
3.170
0.002
0.086
0.363
Configuration [T.Config_22]
0.0083
0.058
0.144
0.885
-0.104
0.121
Configuration [T.Config_23]
0.6167
0.109
5.675
0.000
0.404
0.830
Configuration [T.Config_24]
0.1251
0.081
1.554
0.120
-0.033
0.283
Configuration [T.Config_25]
0.0968
0.032
2.991
0.003
0.033
0.160
Configuration [T.Config_26]
-0.5959
0.180
-3.307
0.001
-0.949
-0.243
Configuration [T.Config_27]
0.6320
0.092
6.849
0.000
0.451
0.813
Configuration [T.Config_28]
0.5921
0.058
10.259
0.000
0.479
0.705
Configuration [T.Config_29]
0.3425
0.154
2.231
0.026
0.042
0.643
Configuration [T.Config_3]
0.2824
0.089
3.170
0.002
0.108
0.457
Configuration [T.Config_30]
-0.2888
0.053
-5.405
0.000
-0.393
-0.184
Configuration [T.Config_31]
-0.0669
0.088
-0.760
0.447
-0.239
0.106
Configuration [T.Config_32]
0.2802
0.063
4.424
0.000
0.156
0.404
Configuration [T.Config_33]
0.5112
0.041
12.613
0.000
0.432
0.591
Configuration [T.Config_34]
0.7608
0.106
7.157
0.000
0.552
0.969
Configuration [T.Config_35]
0.5604
0.080
6.985
0.000
0.403
0.718
Configuration [T.Config_36]
0.5282
0.057
9.192
0.000
0.416
0.641
Configuration [T.Config_37]
-0.0518
0.118
-0.437
0.662
-0.284
0.180
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
40
Section SI.C: Results and statistical analysis of the multivariate linear regression (MLR) model
Configuration [T.Config_38]
0.5113
0.366
1.395
0.163
-0.207
1.229
Configuration [T.Config_39]
0.4866
0.127
3.821
0.000
0.237
0.736
Configuration [T.Config_4]
-0.2546
0.058
-4.381
0.000
-0.368
-0.141
Configuration [T.Config_5]
-0.4323
0.050
-8.563
0.000
-0.531
-0.333
Configuration [T.Config_6]
0.1675
0.064
2.614
0.009
0.042
0.293
Configuration [T.Config_7]
0.9885
0.179
5.520
0.000
0.637
1.339
Configuration [T.Config_8]
-0.4431
0.073
-6.069
0.000
-0.586
-0.300
log(CAP + 1)
0.3920
0.010
40.373
0.000
0.373
0.411
log(LSO + 1)
0.0314
0.007
4.794
0.000
0.019
0.044
log(LSW + 1)
0.2607
0.005
51.910
0.000
0.251
0.271
log(MSO + 1)
0.2232
0.005
40.704
0.000
0.212
0.234
log(MSW + 1)
0.2102
0.006
32.852
0.000
0.198
0.223
log(HSO + 1)
0.1588
0.006
26.461
0.000
0.147
0.171
log(HSW + 1)
-0.0048
0.012
-0.406
0.684
-0.028
0.018
Omnibus:
280.999
Durbin-Watson:
2.017
Prob(Omnibus):
0.000
Jarque-Bera (JB):
350.634
Skew:
-0.326
Prob(JB):
7.26e-77
Kurtosis:
3.587
Cond. No.
258.
Table SI.C4. Regression results for KJF prediction with the MLR model.
Dep. Variable:
log(KJF + 1)
Model:
OLS
Method:
Least Squares
R-squared:
0.837
Adj. R-squared:
0.836
F-statistic:
3704.
Prob (F-statistic):
0.00
Log-Likelihood:
-8168.0
No. Observations:
10922
AIC:
1.642e+04
Df Residuals:
10878
BIC:
1.675e+04
Df Model:
43
Covariance Type:
HAC
coef
std err
z
P>|z|
[0.025
0.975]
Intercept
-0.6000
0.017
-35.501
0.000
-0.633
-0.567
Configuration [T.Config_1]
-0.1208
0.015
-7.935
0.000
-0.151
-0.091
Configuration [T.Config_10]
0.0284
0.032
0.898
0.369
-0.034
0.090
Configuration [T.Config_11]
-0.2395
0.020
-11.698
0.000
-0.280
-0.199
Configuration [T.Config_13]
-0.4837
0.036
-13.474
0.000
-0.554
-0.413
Configuration [T.Config_14]
-0.5217
0.159
-3.281
0.001
-0.833
-0.210
Configuration [T.Config_15]
-0.3123
0.033
-9.453
0.000
-0.377
-0.248
Configuration [T.Config_16]
0.0513
0.057
0.893
0.372
-0.061
0.164
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
41
Section SI.C: Results and statistical analysis of the multivariate linear regression (MLR) model
Configuration [T.Config_17]
-0.4568
0.046
-9.965
0.000
-0.547
-0.367
Configuration [T.Config_18]
-0.1369
0.033
-4.140
0.000
-0.202
-0.072
Configuration [T.Config_2]
-0.2581
0.035
-7.435
0.000
-0.326
-0.190
Configuration [T.Config_20]
0.6336
0.063
10.086
0.000
0.510
0.757
Configuration [T.Config_21]
-0.4329
0.047
-9.263
0.000
-0.524
-0.341
Configuration [T.Config_22]
1.5565
0.026
59.640
0.000
1.505
1.608
Configuration [T.Config_23]
-0.1560
0.073
-2.147
0.032
-0.298
-0.014
Configuration [T.Config_24]
0.4401
0.053
8.300
0.000
0.336
0.544
Configuration [T.Config_25]
-0.3571
0.022
-16.551
0.000
-0.399
-0.315
Configuration [T.Config_26]
-0.7439
0.107
-6.940
0.000
-0.954
-0.534
Configuration [T.Config_27]
-0.0287
0.110
-0.262
0.793
-0.244
0.186
Configuration [T.Config_28]
0.2485
0.098
2.540
0.011
0.057
0.440
Configuration [T.Config_29]
0.5427
0.079
6.872
0.000
0.388
0.697
Configuration [T.Config_3]
-0.2217
0.065
-3.429
0.001
-0.348
-0.095
Configuration [T.Config_30]
-0.4222
0.033
-12.628
0.000
-0.488
-0.357
Configuration [T.Config_31]
-0.7316
0.039
-18.824
0.000
-0.808
-0.655
Configuration [T.Config_32]
-0.4531
0.061
-7.388
0.000
-0.573
-0.333
Configuration [T.Config_33]
-0.0496
0.029
-1.728
0.084
-0.106
0.007
Configuration [T.Config_34]
0.0261
0.105
0.250
0.803
-0.179
0.231
Configuration [T.Config_35]
-0.3327
0.061
-5.481
0.000
-0.452
-0.214
Configuration [T.Config_36]
0.1344
0.060
2.227
0.026
0.016
0.253
Configuration [T.Config_37]
0.0599
0.067
0.891
0.373
-0.072
0.192
Configuration [T.Config_38]
-0.2419
0.172
-1.407
0.159
-0.579
0.095
Configuration [T.Config_39]
0.3052
0.182
1.675
0.094
-0.052
0.662
Configuration [T.Config_4]
-0.2591
0.039
-6.696
0.000
-0.335
-0.183
Configuration [T.Config_5]
-0.3023
0.042
-7.268
0.000
-0.384
-0.221
Configuration [T.Config_6]
-0.1405
0.041
-3.427
0.001
-0.221
-0.060
Configuration [T.Config_7]
0.4759
0.113
4.211
0.000
0.254
0.697
Configuration [T.Config_8]
-0.6683
0.025
-27.166
0.000
-0.717
-0.620
log(CAP + 1)
0.3226
0.008
42.426
0.000
0.308
0.338
log(LSO + 1)
0.1250
0.005
22.802
0.000
0.114
0.136
log(LSW + 1)
0.1609
0.004
41.100
0.000
0.153
0.169
log(MSO + 1)
0.1998
0.004
49.404
0.000
0.192
0.208
log(MSW + 1)
0.1417
0.005
30.277
0.000
0.133
0.151
log(HSO + 1)
0.0812
0.004
18.182
0.000
0.072
0.090
log(HSW + 1)
0.0746
0.009
8.487
0.000
0.057
0.092
Omnibus:
307.127
Durbin-Watson:
Prob(Omnibus):
0.000
Jarque-Bera (JB):
480.099
Skew:
0.276
Prob(JB):
5.60e-105
Kurtosis:
3.867
Cond. No.
258.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
2.015
42
Section SI.C: Results and statistical analysis of the multivariate linear regression (MLR) model
Table SI.C5. Regression results for DGO prediction with the MLR model.
Dep. Variable:
log(DGO + 1)
Model:
OLS
Method:
Least Squares
R-squared:
0.878
Adj. R-squared:
0.877
F-statistic:
3245.
Prob (F-statistic):
0.00
Log-Likelihood:
-10081.
No. Observations:
10922
AIC:
2.025e+04
Df Residuals:
10878
BIC:
2.057e+04
Df Model:
43
Covariance Type:
HAC
coef
std err
z
P>|z|
[0.025
0.975]
Intercept
-0.6353
0.019
-33.446
0.000
-0.672
-0.598
Configuration [T.Config_1]
-0.0532
0.017
-3.217
0.001
-0.086
-0.021
Configuration [T.Config_10]
0.2934
0.031
9.452
0.000
0.233
0.354
Configuration [T.Config_11]
-0.1741
0.023
-7.692
0.000
-0.218
-0.130
Configuration [T.Config_13]
-0.7593
0.066
-11.462
0.000
-0.889
-0.629
Configuration [T.Config_14]
0.4459
0.084
5.332
0.000
0.282
0.610
Configuration [T.Config_15]
-0.0560
0.049
-1.135
0.256
-0.153
0.041
Configuration [T.Config_16]
-0.1889
0.055
-3.419
0.001
-0.297
-0.081
Configuration [T.Config_17]
-0.2232
0.065
-3.425
0.001
-0.351
-0.095
Configuration [T.Config_18]
0.0003
0.034
0.009
0.993
-0.066
0.066
Configuration [T.Config_2]
-0.3399
0.057
-5.978
0.000
-0.451
-0.228
Configuration [T.Config_20]
0.5564
0.089
6.237
0.000
0.382
0.731
Configuration [T.Config_21]
0.1503
0.072
2.078
0.038
0.009
0.292
Configuration [T.Config_22]
0.8556
0.029
29.228
0.000
0.798
0.913
Configuration [T.Config_23]
0.0013
0.077
0.017
0.987
-0.150
0.152
Configuration [T.Config_24]
0.4963
0.059
8.471
0.000
0.381
0.611
Configuration [T.Config_25]
-0.2988
0.029
-10.319
0.000
-0.356
-0.242
Configuration [T.Config_26]
-0.3169
0.143
-2.214
0.027
-0.597
-0.036
Configuration [T.Config_27]
0.3061
0.096
3.191
0.001
0.118
0.494
Configuration [T.Config_28]
0.4641
0.040
11.471
0.000
0.385
0.543
Configuration [T.Config_29]
0.2236
0.129
1.740
0.082
-0.028
0.476
Configuration [T.Config_3]
0.2136
0.109
1.955
0.051
-0.000
0.428
Configuration [T.Config_30]
-0.4128
0.053
-7.734
0.000
-0.517
-0.308
Configuration [T.Config_31]
-0.3430
0.087
-3.950
0.000
-0.513
-0.173
Configuration [T.Config_32]
0.0850
0.055
1.555
0.120
-0.022
0.192
Configuration [T.Config_33]
-0.0026
0.038
-0.068
0.946
-0.077
0.072
Configuration [T.Config_34]
0.0806
0.108
0.743
0.458
-0.132
0.293
Configuration [T.Config_35]
-0.2029
0.059
-3.425
0.001
-0.319
-0.087
Configuration [T.Config_36]
0.3791
0.057
6.625
0.000
0.267
0.491
Configuration [T.Config_37]
-0.5920
0.115
-5.144
0.000
-0.818
-0.366
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
43
Section SI.C: Results and statistical analysis of the multivariate linear regression (MLR) model
Configuration [T.Config_38]
-0.3032
0.685
-0.443
0.658
-1.645
1.038
Configuration [T.Config_39]
0.3242
0.077
4.229
0.000
0.174
0.474
Configuration [T.Config_4]
-0.5547
0.050
-11.011
0.000
-0.653
-0.456
Configuration [T.Config_5]
-0.3474
0.061
-5.729
0.000
-0.466
-0.229
Configuration [T.Config_6]
-0.1927
0.045
-4.238
0.000
-0.282
-0.104
Configuration [T.Config_7]
0.2681
0.139
1.923
0.054
-0.005
0.541
Configuration [T.Config_8]
-0.7317
0.072
-10.215
0.000
-0.872
-0.591
log(CAP + 1)
0.4509
0.009
48.762
0.000
0.433
0.469
log(LSO + 1)
0.0329
0.006
5.069
0.000
0.020
0.046
log(LSW + 1)
0.2578
0.005
54.622
0.000
0.249
0.267
log(MSO + 1)
0.2877
0.005
55.049
0.000
0.277
0.298
log(MSW + 1)
0.2390
0.006
36.777
0.000
0.226
0.252
log(HSO + 1)
0.1360
0.005
24.757
0.000
0.125
0.147
log(HSW + 1)
0.0225
0.015
1.453
0.146
-0.008
0.053
Omnibus:
33.414
Durbin-Watson:
2.011
Prob(Omnibus):
0.000
Jarque-Bera (JB):
36.390
Skew:
0.099
Prob(JB):
1.25e-08
Kurtosis:
3.202
Cond. No.
258.
Table SI.C6. Regression results for HFO prediction with the MLR model.
Dep. Variable:
log(HFO + 1)
Model:
OLS
Method:
Least Squares
R-squared:
0.656
Adj. R-squared:
0.655
F-statistic:
1064.
Prob (F-statistic):
0.00
Log-Likelihood:
-13035.
No. Observations:
10922
AIC:
2.616e+04
Df Residuals:
10878
BIC:
2.648e+04
Df Model:
43
Covariance Type:
HAC
Intercept
coef
std err
z
P>|z|
[0.025
0.975]
-0.5816
0.023
-25.292
0.000
-0.627
-0.537
Configuration [T.Config_1]
0.1263
0.025
5.134
0.000
0.078
0.175
Configuration [T.Config_10]
0.0021
0.042
0.049
0.961
-0.080
0.084
Configuration [T.Config_11]
-0.4291
0.033
-13.083
0.000
-0.493
-0.365
Configuration [T.Config_13]
-1.1194
0.042
-26.744
0.000
-1.201
-1.037
Configuration [T.Config_14]
0.6554
0.093
7.026
0.000
0.473
0.838
Configuration [T.Config_15]
-0.0817
0.067
-1.213
0.225
-0.214
0.050
Configuration [T.Config_16]
-0.8020
0.075
-10.643
0.000
-0.950
-0.654
Configuration [T.Config_17]
-0.7005
0.089
-7.894
0.000
-0.874
-0.527
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
44
Section SI.C: Results and statistical analysis of the multivariate linear regression (MLR) model
Configuration [T.Config_18]
-0.4245
0.045
-9.373
0.000
-0.513
-0.336
Configuration [T.Config_2]
-0.7453
0.048
-15.496
0.000
-0.840
-0.651
Configuration [T.Config_20]
-0.2467
0.162
-1.527
0.127
-0.563
0.070
Configuration [T.Config_21]
-1.1556
0.090
-12.792
0.000
-1.333
-0.979
Configuration [T.Config_22]
0.4615
0.042
11.078
0.000
0.380
0.543
Configuration [T.Config_23]
-0.6237
0.094
-6.640
0.000
-0.808
-0.440
Configuration [T.Config_24]
-1.4601
0.080
-18.342
0.000
-1.616
-1.304
Configuration [T.Config_25]
-1.1318
0.034
-32.902
0.000
-1.199
-1.064
Configuration [T.Config_26]
-0.2048
0.106
-1.926
0.054
-0.413
0.004
Configuration [T.Config_27]
-0.3174
0.159
-2.002
0.045
-0.628
-0.007
Configuration [T.Config_28]
-1.6573
0.111
-14.935
0.000
-1.875
-1.440
Configuration [T.Config_29]
-0.4312
0.177
-2.443
0.015
-0.777
-0.085
Configuration [T.Config_3]
-0.4578
0.141
-3.235
0.001
-0.735
-0.180
Configuration [T.Config_30]
-1.4906
0.054
-27.715
0.000
-1.596
-1.385
Configuration [T.Config_31]
-1.6854
0.068
-24.830
0.000
-1.818
-1.552
Configuration [T.Config_32]
-0.7913
0.089
-8.887
0.000
-0.966
-0.617
Configuration [T.Config_33]
-1.5271
0.053
-28.836
0.000
-1.631
-1.423
Configuration [T.Config_34]
-2.6699
0.170
-15.670
0.000
-3.004
-2.336
Configuration [T.Config_35]
-2.2686
0.067
-33.981
0.000
-2.399
-2.138
Configuration [T.Config_36]
-1.8854
0.084
-22.388
0.000
-2.050
-1.720
Configuration [T.Config_37]
-0.6738
0.106
-6.341
0.000
-0.882
-0.466
Configuration [T.Config_38]
-1.3440
0.718
-1.872
0.061
-2.751
0.063
Configuration [T.Config_39]
-2.2781
0.123
-18.594
0.000
-2.518
-2.038
Configuration [T.Config_4]
-0.8567
0.038
-22.757
0.000
-0.931
-0.783
Configuration [T.Config_5]
-0.2387
0.084
-2.857
0.004
-0.402
-0.075
Configuration [T.Config_6]
-0.4608
0.052
-8.828
0.000
-0.563
-0.359
Configuration [T.Config_7]
-0.4753
0.101
-4.686
0.000
-0.674
-0.277
Configuration [T.Config_8]
-0.4922
0.149
-3.304
0.001
-0.784
-0.200
log(CAP + 1)
0.4748
0.011
42.908
0.000
0.453
0.497
log(LSO + 1)
0.0188
0.008
2.286
0.022
0.003
0.035
log(LSW + 1)
0.0459
0.006
7.660
0.000
0.034
0.058
log(MSO + 1)
0.2631
0.006
42.592
0.000
0.251
0.275
log(MSW + 1)
0.1317
0.007
17.746
0.000
0.117
0.146
log(HSO + 1)
0.0691
0.007
9.304
0.000
0.055
0.084
log(HSW + 1)
-0.0416
0.016
-2.667
0.008
-0.072
-0.011
Omnibus:
123.340
Durbin-Watson:
1.964
Prob(Omnibus):
0.000
Jarque-Bera (JB):
140.852
Skew:
-0.212
Prob(JB):
2.60e-31
Kurtosis:
3.361
Cond. No.
258.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
45
Notes
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
46
Notes
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
47
Notes
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
48
Notes
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
49
About the Author
Dr. Evar Umeozor is a Research Fellow at KAPSARC leading integrated
oil and gas value-chain research, with a focus on the downstream
segment. Dr. Umeozor is a process and energy systems engineering
expert with multidisciplinary research and industry experience in technical
assessments, life cycle cost and emission analyses, and the application
of the systems approach to optimal policy design considering multiple
objectives and constraints.
He holds a master’s degree in chemical engineering from Imperial
College London, United Kingdom, and a doctoral degree in energy and
environment systems engineering from the University of Calgary, Canada.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
50
About the Project
This study was conducted under the KAPSARC Oil Value-Chain Analyzer (KOVA) project,
which develops data, models, and analytics tools to gain insights into the impacts of
business and government policies on the economic, environmental and energy efficiencies
of the downstream oil and gas sector, including the integral effects on the midstream
and upstream sectors in Saudi Arabia and beyond. The objective of the KOVA project is
to deploy integrated systems analysis approaches for identifying optimal policy design
options that satisfy the performance metrics and targets of stakeholders in the energy
ecosystem.
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
51
www.kapsarc.org
Global Crude Oil Refinery Models: Parametric and Nonparametric Methods
52